Development of a Scalable Bias Mitigation Framework for AI-Based Consumer Credit Scoring in Regulated U.S. Lending Models
Applied IT & Engineering · 01.01.2024 · 2 (1)
DOI: 10.25163/engineering.2110481 ↗Kurz gefasst
Fachartikel in der Zeitschrift Applied IT & Engineering, online erschienen am 01.01.2024. Die Arbeit wurde bisher 1-mal zitiert (Crossref).
Abstract (Originalsprache)
Background
The U.S. lending industry operates with AI-based consumer credit scoring systems but these systems maintain existing prejudices which stem from personal characteristics and monetary situations. The creation of bias reduction methods which work on a large scale stands as a fundamental requirement to protect fairness while keeping prediction accuracy intact.
Methods
The research gathered data from 295 participants who provided their demographic information through age and gender and education level and occupation and marital status and their financial behavior through debt-to-income ratio and repayment timeliness and savings and loan approval outcomes. Pearson correlation analysis examined relationships among variables, and association analysis using standardized Beta coefficients, p-values, and chi-square (χ²) statistics assessed predictor influence on lending outcomes.
Results
Age showed positive correlations with education (r = 0.38, p = 0.01) and repayment timeliness (r = 0.30, p = 0.01). The analysis shows that education and savings function as main predictors but gender and marital status do not show significant effects after mitigation. The AI model showed a decrease in performance metrics with accuracy dropping from 82.1 to 80.5 and precision from 78.3 to 77.9 and recall from 75.6 to 75.0 but the fairness metrics demonstrated substantial improvement through decreased demographic parity and equal opportunity differences.
Conclusion
The bias mitigation framework achieves both fairness improvement and predictive reliability maintenance for AI-based credit scoring systems. The mixture of demographic data with financial minutiae and interactive patterns allows fair and see-through loaning selections under U.S. regulatory values.
Zitieren
APA
(2024). Development of a Scalable Bias Mitigation Framework for AI-Based Consumer Credit Scoring in Regulated U.S. Lending Models. Applied IT & Engineering, 2(1). https://doi.org/10.25163/engineering.2110481
Vancouver
Development of a Scalable Bias Mitigation Framework for AI-Based Consumer Credit Scoring in Regulated U.S. Lending Models. Applied IT & Engineering. 2024;2(1). doi:10.25163/engineering.2110481
BibTeX
@article{euvimed2024,
title = {Development of a Scalable Bias Mitigation Framework for AI-Based Consumer Credit Scoring in Regulated U.S. Lending Models},
journal = {Applied IT & Engineering},
year = {2024},
volume = {2},
number = {1},
doi = {10.25163/engineering.2110481}
}
EUVIMED